Comparison

Humalike vs Thinking Machines.

Thinking Machines bakes interaction into the foundation model. Humalike exposes it as composable primitives above any LLM — usable today.

Them

Thinking Machines

Frontier AI lab

Mira Murati's lab, $2B-funded. TML-Interaction-Small (May 2026 research preview) — 276B-parameter MoE (12B active), 0.4s response, full-duplex audio/video/text. The interaction loop lives in the model weights.

Us

Humalike

Behavioral infrastructure

Composable behavioral APIs — turn-taking, norms, persona, memory, theory of mind, social signals — wrapped around whatever LLM you ship. Closed beta with design partners today.

The core differences.

No feature-checklist arms race. Just the dimensions that decide which one fits — with the honest call on each.

Generally available product

Thinking Machines

Closed research preview.

Humalike

Closed beta with design partners.

Builds foundation models

Thinking Machines

Humalike

Model-agnostic

Thinking Machines

Their own interaction model.

Humalike

Bring any LLM.

Composable above any LLM

Thinking Machines

Interaction is in the weights.

Humalike

Pick the APIs you need.

Multi-human rooms

Thinking Machines

Designed for 1:1 with the model.

Humalike

Multi-party as a first-class case.

Persistent memory across sessions

Thinking Machines

Humalike

Social signals (typing, edits, pauses)

Thinking Machines

Humalike